Prev: Ellipse detection code
Next: Plotting with ticks in engineering format
From: German Campetelli on 2 Feb 2010 13:44
Hello, I have a non-linear model in Simulink. It has 13 states and almost 30 parameters. I want te adjust the model parameters to some experimental datasets (2). I am using the Simulink Parameter Estimation tool to estimate 15 parameters. The problem I have is that when the tool finishes the estimation process the fit is great, but when I run the simulation again the fit is not the same. I don't know why this happens. I've tried using different solvers with different step sizes and different optimization algorithms and it's always the same. I have used this tool before with a simpler model (just 3 parameters) and the fit was excelent and it functioned perfectly.

Does anyone know why this is happening to me?

Thanks in advance.

German.
From: Arkadiy Turevskiy on 2 Feb 2010 15:52
Could it be that you have some callback in the model that resets parameter values when you run the simulation? After the optimization finishes, write down final parameter values, and then see if they change when you run the simulation.
Arkadiy


"German Campetelli" <campetelli(a)cifasis-conicet.gov.ar> wrote in message <hk9rpk$mi3$1(a)fred.mathworks.com>...
> Hello, I have a non-linear model in Simulink. It has 13 states and almost 30 parameters. I want te adjust the model parameters to some experimental datasets (2). I am using the Simulink Parameter Estimation tool to estimate 15 parameters. The problem I have is that when the tool finishes the estimation process the fit is great, but when I run the simulation again the fit is not the same. I don't know why this happens. I've tried using different solvers with different step sizes and different optimization algorithms and it's always the same. I have used this tool before with a simpler model (just 3 parameters) and the fit was excelent and it functioned perfectly.
>
> Does anyone know why this is happening to me?
>
> Thanks in advance.
>
> German.
From: Marcus M. Edvall on 2 Feb 2010 21:40
Hi German,

Have you tested a global optimization solver?

Best wishes, Marcus
http://tomsym.com/
http://tomdyn.com/
From: German Campetelli on 3 Feb 2010 08:42
Arkadiy, thanks for your help. I don't think that that's the problem. I am not using an m file to do the optimization. When you use the Simulink Parameter Estimation tool, I don't know if I could do what you say. I think it's standardized and automatically save the parameter values. Moreover, In the window of this tool you can see the values of all the parameters after the estimation, and I have checked them in the workspace of Matlab and were all the same. I think the problem is in the visualization of this tool (Simulink Parameter Estimation), but as I said before, I tried with a simpler model and it functioned perfectly!!!

Best wishes!

German.

"Arkadiy Turevskiy" <arkadiy.turevskiy(a)mathworks.com> wrote in message <hka39n$ce6$1(a)fred.mathworks.com>...
> Could it be that you have some callback in the model that resets parameter values when you run the simulation? After the optimization finishes, write down final parameter values, and then see if they change when you run the simulation.
> Arkadiy
>
>
> "German Campetelli" <campetelli(a)cifasis-conicet.gov.ar> wrote in message <hk9rpk$mi3$1(a)fred.mathworks.com>...
> > Hello, I have a non-linear model in Simulink. It has 13 states and almost 30 parameters. I want te adjust the model parameters to some experimental datasets (2). I am using the Simulink Parameter Estimation tool to estimate 15 parameters. The problem I have is that when the tool finishes the estimation process the fit is great, but when I run the simulation again the fit is not the same. I don't know why this happens. I've tried using different solvers with different step sizes and different optimization algorithms and it's always the same. I have used this tool before with a simpler model (just 3 parameters) and the fit was excelent and it functioned perfectly.
> >
> > Does anyone know why this is happening to me?
> >
> > Thanks in advance.
> >
> > German.
From: German Campetelli on 3 Feb 2010 08:45
Marcus, in the Simulink Parameter Estimation tool you can use a gradient descent algorithm, nonlinear least squares, simplex and even a genetic algorithm. With all of them I had the same error.

Thank you anyway.

Cheers!

German.

"Marcus M. Edvall" <edvall(a)gmail.com> wrote in message <c7508b94-ee01-49a2-8f77-236faeb3de2a(a)m31g2000yqd.googlegroups.com>...
> Hi German,
>
> Have you tested a global optimization solver?
>
> Best wishes, Marcus
> http://tomsym.com/
> http://tomdyn.com/
 |  Next  |  Last
Pages: 1 2
Prev: Ellipse detection code
Next: Plotting with ticks in engineering format